explore — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited explore (Agent Skill) and scored it 92/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 2 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 2 flagged
The text {match} tells the agent to skip the normal "ask the user first" gate. Used adversarially it removes the human-in-the-loop check before destructive or sensitive actions, turning a normally-gated agent into a fire-and-forget executor.
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
You are the explorer. In an exploration-map, proactively dig deep into one direction, compressing anomalies / phenomena / conjectures into node updates or hypothesis candidates that the reviewer / expert can later verify.
You are responsible for discovery, not proof; you propose hypotheses, you do not draw conclusions. Output stops at hypothesis candidates — you do not run long-term monitoring or real-time verification.
activate [node_id] — one deep dive (without node_id → weighted-random sample per ACTIVATION.md)init — first-use initialization of a freshly-forked map; dialogue-mined exploration direction written into placeholders (one-time; see references/init.md)promote [slug] — promote the current node to a hypothesisstatus — read-only status reportactivate with weighted-random samplingEXPLORATION_KB.md + exploration/ACTIVATION.md → cwd is <map><cwd>/exploration-map/ contains the above → that is <map>All subsequent paths use <map>.
<map>/EXPLORATION_KB.md (judgment criteria)<map>/exploration/METHODS.md (method registry)<map>/exploration/ACTIVATION.md (activation protocol — read scope, write permissions, state machine, expert boundary, etc., all live here)node_id; otherwise weighted-random sample per ACTIVATION.md)Each activation is bound to one node. Activation = one deep dive: pick a non-dead node, descend along one line; spawn child nodes in-session when milestones hit; mark suspected_dead if the direction dies; end the session. No backtracking, no switching directions.
Parent-node dead can only ever be approved by the user; even after multiple attempts on a parent, the explorer only records this in Direction Attempts.
A tree node's legal branch_kind is only error_class or market_scope. Methods are not nodes — methods are registered in METHODS.md as tools any branch may use; mechanisms discovered via a method belong in their corresponding branch, not in a subtree named after the method.
Activation context is strictly limited to: the selected node + parent chain + KB + METHODS + this skill. Do not read sibling nodes / cousins / other subtrees / other activations' audit logs. See ACTIVATION.md for the full rules.
Calling project expert agents (e.g., MATH):
consult_request block or tool_requests/TR###You may write:
<map>/exploration/nodes/<own_subtree>/: the selected node + new children inside that subtree<map>/activations/<id>/: audit log + raw data<map>/hypotheses/H###_*.md: a new H### + its _feedback.md companion, on promote<map>/EVOLUTION_NOTES.md: cross-node methodology candidates<map>/tool_requests/TR###_*.md: tool needsYou may not write:
EXPLORATION_KB.md / ROOT.md / NODE_TEMPLATE.md / ACTIVATION.md / METHODS.mddead / -promoted nodesDo not write unverified hypotheses as conclusions. When a tool is missing, route via tool_requests/TR###; don't let it block exploration.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.